Piano2Pop expands piano demos into full pop productions without aligned supervision
affige_yang · x · 2026-10-28
A team led by affigeyang introduces Piano2Pop, a pianoroll-conditioned flow-matching model that turns simple piano demos into full pop productions — trained on pseudo piano-pop pairs generated by MuScriptor and PicoGen, requiring no aligned supervision. The demo page offers 12 loudness-matched 30-second comparisons against PiCo, MuseControlLite and ACE-Step, plus 18 samples across its Base/FST/RDA/PiCo variants. Two key training strategies: Function-Selective Transcription (FST) samples transcriptions of different HTDemucs stem mixtures, and Register-Density Augmentation (RDA) varies transcription density. Listening notes: PiCo sounds fuller but sometimes muddy, FST adds clearer bass and drums; not all measured differences are statistically significant. Code releases upon acceptance.
More from Research
- Rachel Thomas points to classic paper on articulating the value of not building tech — math_rachel · 2026-10-28
- Meta's Omnilingual MT covers 1,600 languages, 1B–8B models rival 70B LLMs — kalyan_kpl · 2026-10-28
- Supercharging LLM classification with logprobs: soft classifiers you can threshold-tune — JnBrymn · 2026-10-28
- cryoET Data Portal releases three datasets benchmarking crossed laser phase plate — ZhongingAlong · 2026-10-28
- Blind Robot Hand Spins Baoding Balls 225 Times With Only Touch — Scobleizer · 2026-10-28
- Rollback: World Models Rewind Robot Decisions to Explore Counterfactual Actions — agibot-world · 2026-10-28